Dataloader overrides Transforms to a boolean

[leonelluiscorado]

Hi! I'm trying to build a custom dataset inherited from RasterDataset.
I only need an additional argument which is a .csv file with the targets, while everything almost remains the same:

class MyCustomDataset(RasterDataset):
    filename_glob = "*_sen2.tif"
    #filename_regex = r"\d+_(?P<band>B[\d])_sen2"
    is_image = True
    separate_files = False
    all_bands = ["B1", "B11", "B12", "B2", "B3", "B4", "B5", "B6", "B7", "B8", "B8a", "B9"]
    rgb_bands = ["B4", "B3", "B2"]
    
    targets = None
    
    def __init__(self, root="data", crs=None, res=None, transforms=None, cache=True, targets_csv=None):
        self.root = root
        super().__init__(root, crs, res, transforms, cache)
        
        if targets_csv is not None:
            self.targets = pd.read_csv(targets_csv, index_col=0)
            self.targets = torch.Tensor(self.targets["rh98"])

    def plot(self, sample):
        # Find the correct band index order
        rgb_indices = []
        for band in self.rgb_bands:
            rgb_indices.append(self.all_bands.index(band))

        # Reorder and rescale the image
        image = sample["image"][rgb_indices].permute(1, 2, 0)

        # Plot the image
        fig, ax = plt.subplots()
        ax.imshow(image)

        return fig

I followed the tutorial on the TorchGeo docs "Custom Datasets". When I try to instantiate a Dataloader and iterate through the samples:

custom = MyCustomDataset(data_dir, targets_csv=targets_csv)

sampler = PreChippedGeoSampler(dataset)   # Ds Already in Chips
dataloader = DataLoader(dataset, sampler=sampler, collate_fn=stack_samples)

At this line:

batch = next(iter(dataloader))

It gives me the following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[101], line 1
----> 1 batch = next(dataloader)

File ~/rh98-estimation-dl/torchgeo/env/lib/python3.8/site-packages/torch/utils/data/dataloader.py:628, in _BaseDataLoaderIter.__next__(self)
    625 if self._sampler_iter is None:
    626     # TODO(https://github.com/pytorch/pytorch/issues/76750)
    627     self._reset()  # type: ignore[call-arg]
--> 628 data = self._next_data()
    629 self._num_yielded += 1
    630 if self._dataset_kind == _DatasetKind.Iterable and \
    631         self._IterableDataset_len_called is not None and \
    632         self._num_yielded > self._IterableDataset_len_called:

File ~/rh98-estimation-dl/torchgeo/env/lib/python3.8/site-packages/torch/utils/data/dataloader.py:671, in _SingleProcessDataLoaderIter._next_data(self)
    669 def _next_data(self):
    670     index = self._next_index()  # may raise StopIteration
--> 671     data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
    672     if self._pin_memory:
    673         data = _utils.pin_memory.pin_memory(data, self._pin_memory_device)

File ~/rh98-estimation-dl/torchgeo/env/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py:58, in _MapDatasetFetcher.fetch(self, possibly_batched_index)
     56         data = self.dataset.__getitems__(possibly_batched_index)
     57     else:
---> 58         data = [self.dataset[idx] for idx in possibly_batched_index]
     59 else:
     60     data = self.dataset[possibly_batched_index]

File ~/rh98-estimation-dl/torchgeo/env/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py:58, in <listcomp>(.0)
     56         data = self.dataset.__getitems__(possibly_batched_index)
     57     else:
---> 58         data = [self.dataset[idx] for idx in possibly_batched_index]
     59 else:
     60     data = self.dataset[possibly_batched_index]

File ~/rh98-estimation-dl/torchgeo/env/lib/python3.8/site-packages/torchgeo/datasets/geo.py:436, in RasterDataset.__getitem__(self, query)
    433     sample["mask"] = data.long()
    435 if self.transforms is not None:
--> 436     sample = self.transforms(sample)
    438 return sample

TypeError: 'bool' object is not callable

Entering line 436 means that the transforms variable is not None, which means it overrides to something, which is a 'bool' object.
I tried searching forums on PyTorch and every answer points to the person calling transforms as a bool, while in my case, I say that transforms=None .

In the RasterDataset class, the transforms argument is also None.

I also tried defining transforms=None when instantiating an object of my class, but the same problem occurs.
Also tried updating torchgeo.

Is there something I'm missing when I'm creating my custom class? Why does the transforms variable override to a boolean?

I'm really thankful for any tips and answers!

[leonelluiscorado]

I forgot the transforms=transforms on the :

super().__init__(root, crs, res, transforms, cache)

changed to:

super().__init__(root, crs, res, transforms=transforms, cache=cache)

[adamjstewart]

Note that you may also need to override __getitem__ like so:

def __getitem__(self, query: BoundingBox) -> Dict[str, Any]:
    sample = super()[query]
    # idk how you plan on matching a particular geographic query to a target label
    target = self.targets[magic]
    sample["label"] = target
    return sample

[leonelluiscorado]

Hi! Thank you very much for the reply.
Matching a geographic query to a target label is simple, I'm doing it with an index. Every image is named after an index from 0 to N, and I have a csv file with 0 to N targets.

Can I add the idx (index) to the getitem function?

[adamjstewart]

If you aren't using the geographic metadata and only want to sample by an integer index, you're better off using NonGeoDataset instead of GeoDataset. Then the dataset is basically identical to the kind of dataset you would write in torchvision. That's how most of our datasets are implemented so there should be plenty of examples to look at.